Google Open-Sources PointDiT: Minimalist ViT for Monocular 3D Reconstruction
CSProfKGD · x · 2026-08-25
Google Research has released PointDiT (ICML 2026), a pixel-space Diffusion Transformer for monocular geometry estimation. The project is fully open-sourced, including code, pre-trained models, and the full training/evaluation pipeline.
PointDiT features a minimalist design: operating directly on raw patches to estimate dense 3D point maps using a plain ViT. It eliminates the need for hybrid ViT+Conv architectures, lossy VAEs, or complex training losses, relying instead on conditioning on frozen pre-trained DINOv3 image tokens.
More from Research
- Does Moderna's cancer vaccine use AI? ML likely helps at two key stages — HarperSCarroll · 2026-08-25
- One week plan: Study Kernels and Transformers docs, reproduce code, and fix bugs — ariG23498 · 2026-08-25
- New blog post critiques weaknesses in OpenAI's Monitorability Evals — TuhinChakr · 2026-08-25
- AI field sees massive progress but unsolved questions are growing faster — littmath · 2026-08-25
- Deep learning solves perovskite kinetics inverse problem, revealing key parameters — bravo_abad · 2026-08-25
- Open Source AI Eval Skills Guide Coding Agents — petergyang · 2026-08-25